Home

About the model

The Joint Research Centre (JRC) of the European Commission wanted to develop projections of how built-up land and population levels will likely change locally across the globe. The first objective of this work was to provide the UN Population Division with projections of the population by Degree of Urbanisation up to 2100. The UN Population Division would like to include these global projections in their World Urbanization Prospects publication foreseen for early 2025. The second objective is to develop a flexible and efficient model that can integrate other population projections and global SSP scenarios. These projections must be done spatially explicitly and on a fine spatial resolution (entailing 1 km2 grid cells). Given the complexity of the task, this work required a dedicated GIS-based platform that allows users that have received some training to create, recreate, and adapt projections in a computationally efficient way. In 2024, the JRC and Object Vision developed this model in the GeoDMS software.

How does the model work?

The model downscales exogenously defined additional built-up land and population change on 1km2 grid cells globally (see Geography). In the reference application, the exogenous inputs are defined for a custom set of regional boundaries that are called 'functional areas'. These boundaries are described in more detail in the section define subnational units for projections. Grid cells are included that are maximally 5km away from any grid cell that contains some built-up development in 2020, that are within country borders, and that are not fully covered by water or ice.

Built-up allocation
Built-up is downscaled using functions based on empirically observed built-up presence and built-up development. Built-up is defined as the fraction of land covered by buildings; building heights are not taken into account. The approach to model built-up development is based on four premises:

1. mirroring GHSL assumptions, built-up fractions cannot be reduced, so that only additional built-up fractions are distributed;
2. locations differ in suitability, and more suitable locations are more likely to receive additional built-up;
3. from GHSL built-up grids follows that 60% is a realistic cut-off for built-up development in a grid cell;
4. over time, the statistical distribution of built-up values is remarkably resilient, maintaining similarity to a power-law distribution even in territories with considerable built-up development.

This is done by a linear downscaling of additional land over a grid map of values that combines the locational suitability of every grid cell, a balancing factor, and the expected additional built-up given the prior level of built-up development in the grid cell. Locational suitability is described as a logit probability that is estimated using calibration routines. Expected additional built-up is described through ET-functions ('Expected Top-ups') that in turn were estimated from historical decennial built-up development.

Population allocation
Population is defined as a discrete number of people, without discerning sex or age. Population changes are modelled with universal rules across the entire territory. Global ground-truth population data are unavailable so a statistical model of population change is unobtainable. By necessity, the implemented rules, therefore, reflect coarse assumptions on how population relates to suitabilities for built-up development and responds to additional built-up land. Expected exogenous population changes are thus downscaled using modelled built-up changes as a support variable. In addition to population changes, a parametrized share of the region's population is re-allocated by the model to model internal migration. The population model consists of three steps:

1. first, a population pool is formed that represents all people that are allocated in one iteration a region; then,
2. part of the population from that pool is allocated to new built-up development; and finally,
3. the remainder is distributed while taking into account local built-up suitabilities and population change limits.

The latter population change limits indicate maximum decennial population changes observed in GHS population layers, depending on whether a grid cell already contains population, and whether the grid cell is the most populous in its environment.

The model mechanics section describes the model's inner workings in more detail.

How to use the model

The model is based on open data and the open-source GeoDMS platform. The modelling process requires up to 120Gb of memory for large continents. System requirements are a recent Windows machine with at least 64Gb RAM (or a very large page file). For development purposes, the model is typically ran for six continents with fixed parameters, but the application allows for substantial modification and customisation. For detailed instructions on how to obtain the model, set it up, and run it, see our tutorial page.